fix(nebius): refresh Token Factory integration

[demianarc]

Ticket

N/A — this refreshes the existing Nebius Token Factory integration; it does not add a new provider.

Component/Service

• Web (Frontend)
• Jawn (Backend)
• Worker (Proxy)
• Bifrost (Marketing)
• AI Gateway
• Packages
• Infrastructure/Docker
• Documentation

Type of Change

• Bug fix
• New feature
• Breaking change
• Documentation update
• Performance improvement
• Refactoring

Deployment Notes

• No special deployment steps required
• Database migrations need to run
• Environment variable changes required
• Coordination with other teams needed

Screenshots / Demos

Not applicable (docs and provider behavior only).

Extra Notes

• The existing llama-3.3-70b-instruct:nebius model ID and $0.13/$0.40 per-million-token pricing were re-verified against the public Token Factory catalog on 2026-08-13, so this PR adds regression coverage instead of changing correct registry data.
• The Image Generations example now uses an explicit current-model placeholder because no image-generation model is present in the public catalog at the time of this update. The route remains covered as transparent legacy-proxy behavior.
• This does not claim direct Token Factory Responses API support. Helicone's AI Gateway Nebius provider targets Chat Completions; Helicone may translate Responses-format input to that endpoint.

Context

The existing integration page used retired model examples, omitted the required Helicone-Auth header from both runnable requests, and did not distinguish the transparent legacy proxy from the chat-only AI Gateway provider mapping. The Nebius provider also ignored standard OpenAI-compatible { "error": { "message": ... } } error bodies.

This change:

• updates the chat example to the currently active meta-llama/Llama-3.3-70B-Instruct model;
• removes the retired hard-coded image model while preserving the documented image route;
• adds the missing Helicone authentication header;
• documents the Chat Completions / Responses boundary;
• accepts both Token Factory detail errors and OpenAI-compatible nested errors;
• adds request, pricing, error, and chat/image path-preservation regression tests.

Tests

• yarn jest --config packages/jest.config.ts --runInBand packages/__tests__/cost/providers/nebius.test.ts (5 passed)
• yarn workspace helicone-worker test:run --config test/gateway/vitest.config.mts test/gateway/nebius-routes.spec.ts (3 passed)
• yarn workspace helicone-worker lint (passed; existing warnings only)
• yarn prettier --check docs/getting-started/integration-method/nebius.mdx packages/cost/models/providers/nebius.ts packages/__tests__/cost/providers/nebius.test.ts worker/test/gateway/nebius-routes.spec.ts worker/test/gateway/vitest.config.mts worker/vitest.config.mts

[vercel]

@demianarc is attempting to deploy a commit to the Helicone Team on Vercel.

A member of the Team first needs to authorize it.